How is `/summarize` supposed to work?

Support Bug Reports
1

I’ve noticed the behavior of the /summarize is a little different than other slash-commands. It takes action immediately when selected (not when submitting the prompt, but the moment it is chosen from the /-menu).

Can the team clarify what it is actually doing? Perhaps update the docs with more detail here: Summarization | Cursor Docs ?

  1. I’ve noticed /summarize triggers model work. Shouldn’t this be passive? Is this a bug? Does the model prompt itself and think it is supposed to take action? If this slash command is a prompt, perhaps it needs an agent note to just read it and not change code or investigate any of the contents?

  2. If I type a prompt, then somewhere in the prompt (top, bottom, where-ever) and add the /summarize command, it immediately submits the prompt. What is the effect now? Does it summarize and then apply my prompt?

  3. Does it matter what model I am using when I run the command? For example, if I am using GPT-5 (High) and I want to summarize more quickly, can I switch to Composer or Grok fast? Is the context the current agent has somewhat trapped with the model or can switching models be done freely? (This is perhaps a more general question about using agents that has been on my mind, but also relevant for this topic.

Perhaps this should be a menu item instead of a slash-command to sidestep some of this confusion?

I feel if I can understand this better, I can have more focused and effective results from the agent.

1 Like
2

Now that I am paying attention to this. I noticed an issue when switching models. Does this reasoning make sense? What I did…

  1. I was using GPT-5 High for some time and the context was getting quite full and results seemed to be getting worse
  2. In a new prompt, I switched to Haiku, typed /sum and chose /summarize from the auto-complete and hit enter
  3. It started to plan next moves for some time. Long enough that I thought maybe Cursor failed to switch the model away from GPT-5 High! Then I noticed the context circle was totally full (GPT-5 has 272k, Haiku only has 200k).

…I stopped the prompt, changed it to GPT-5, let it summarize, and then continued working.

Is there an issue switching between models when the new model can’t fit the current agent context in the new model’s window perhaps? Not enough room to summarize?

3

There is definitely some buggy behavior where the model does try and do work when /summarize is run. I ran it and since it executes upon auto-complete selection it just ships an empty prompt and when summarizing is done, the model gets to take its turn…

Screenshot 2025-11-11 at 4.54.41 PM
Screenshot 2025-11-11 at 4.54.41 PM1850×322 28.1 KB

I flagged this post to be promoted to a bug report. I am using:

Version: 2.0.69 (Universal)
VSCode Version: 1.99.3
Commit: 63fcac100bd5d5749f2a98aa47d65f6eca61db30
Date: 2025-11-07T18:21:29.650Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 24.6.0
3 Likes
Gemini 3.0 Pro - Out Now!
4

Bumping for team reply. I think these questions are fair clarification requests?

1 Like
5

Agreed, this is clearly a bug and it’s extremely annoying. Best workaround I have found is to run the command, then stop generation as soon as it says it’s summarized, and send a new message from there.

6

You can just call the command in an already written prompt, and not when you write it :eyes: